Mortgage Documents · API

Upload a document. Get structured data back.

Built for Dutch mortgage documents. The API returns the relevant fields in a consistent format, each with its own confidence score. Seven document types are available.

Talk to us See the modules
PDF
Uploading the PDF.
 reads · checks · structures
Done. 34 fields, each with a confidence score.
Already?
Ready.
How it works

A straightforward three-step flow.

  1. 01

    Upload

    One call with the document and its type. You get a document ID back immediately.

  2. 02

    Wait a moment

    Processing runs in the background. The API takes the time needed to check the result before returning it.

  3. 03

    Collect

    Poll with the ID and receive every field as a name, a value and a confidence.

POST /documents/DocumentsSend
document-type: ls
document-data: loonstrook.pdf

→ { "DocumentID": "f47ac10b-58cc-4372-a567-0e02b2c3d479" }

GET  /documents/GetResults?document-id=f47ac10b-…

→ {
    "Status": "Finished",
    "DocumentType": "ls",
    "FlowData": [
      { "name": "wn_naam", "value": "De Vries", "confidence": 0.98 },
      { "name": "wg_naam", "value": "Voorbeeld B.V.", "confidence": 0.97 },
      { "name": "deeltijdfactor", "value": 0.8, "confidence": 0.99 },
      { "name": "salaris_regulier", "value": 4667, "confidence": 0.97 },
      { "name": "netto_uitbetaald", "value": 3892.5, "confidence": 0.99 }
    ]
  }
Trimmed example. Field names and formats are documented per module.
Modules

Seven document types, each with its own module.

The extraction approach varies by document type. Standardised forms use layout-aware models; documents with many layouts are read by several language models and compared. The result has the same simple shape: a flat list of fields.

01

Employer declaration

Werkgeversverklaring, NHG model
wgv74 fields
Filled in with a blue pen. Signed with a black one. And something crossed out.
Two ink colours. One strikethrough. Both flagged.
You noticed?
Fields 73 and 74.
What goes in

A scanned or digital employer declaration, filled in by hand or typed.

What comes out

Employee and employer details, contract type, probation, reorganisation, the full income table, private loans and wage garnishment, the signature block and the NHG model number.

Worth noticing

The module also performs two visual checks: whether there are strikethroughs or corrections, and whether the form was filled in with more than one ink colour.

02

Identity document

Passport or identity card
legi17 fields
Passport scan, as requested.
Black and white. And the page with the perforated number is missing.
I’ll ask for a new one.
What goes in

A scan of a passport or identity card, front and back.

What comes out

Name, date and place of birth, nationality as an ISO code, document number, issue and expiry date, issuing authority.

Worth noticing

Whether the citizen service number is visible, whether the scan is in colour, and for passports whether the opposite page with the perforated document number is included. The machine-readable zone is verified with its own check digits.

03

Purchase agreement

Koopovereenkomst
kovk43 fields
€ 407.000 vierhonderdzevenduizend
Purchase price € 407.000. In words: vierhonderdzevenduizend.
Digits and words agree. Confidence 0.99.
You read the words too?
Always.
What goes in

A signed purchase agreement for a home, often twenty pages or more.

What comes out

Both buyers, the property, purchase price, movable goods, completion date, notary, bank guarantee and the financing condition.

Worth noticing

Amounts appear twice in a Dutch purchase agreement: in digits and written out. Both are extracted and compared, and the comparison feeds the confidence score.

04

Payslip

Loonstrook
ls34 fields
Every employer has its own payslip layout.
One consistent set of 34 fields.
What goes in

A payslip from any employer or payroll provider, in any layout.

What comes out

Employee and employer, IBAN, hours and part-time factor, contract flags, pay period, salary components, pension contribution, end-of-year bonus, thirteenth month, individual choice budget and net paid.

Worth noticing

Payslips are the least standardised document in the set. Several models read every payslip independently and vote on each field. What they agree on is then confirmed against the text of the document.

05

Appraisal report

Taxatierapport
taxrap112 fields
Sixty pages.
Energy label on page 12. Foundation on page 58. 112 fields.
What goes in

A residential appraisal report, typically fifty pages or more.

What comes out

Client, property, market value, valuation date, leasehold, homeowners’ association, energy label, a sustainability block covering insulation, heating, energy generation and improvement potential, and the foundation report.

Worth noticing

Fields only appear where they apply: leasehold details for leasehold properties, association details for apartments. The sustainability block is the newest part and is still being refined.

06

DUO student loan statement

Overzicht studieschuld
duo51 fields
Student debt?
Two loans. Balance, interest and term for each.
What goes in

A DUO statement as a PDF, a screenshot or a forwarded e-mail.

What comes out

Borrower details, total balance, and for each of up to five separate loans the type, start and current balance, interest, remaining term and repayment rules.

Worth noticing

A statement that explicitly says there is no student debt is recognised as such. That is different from a statement where the loan fields are simply empty.

07

Chamber of Commerce extract

KVK-uittreksel
kvk29 fields
Who is allowed to sign for this company?
Three persons, each with their role.
What goes in

A KVK extract for a company or a sole proprietorship.

What comes out

Trade name, legal form, KVK and RSIN numbers, SBI codes, start date, shareholders and directors, and up to five natural persons with their role.

Worth noticing

Roles come back as stated on the extract: owner, partner, shareholder, director. That is what a lender needs in order to know who may sign.

What makes it reliable

Extract first. Verify. Normalise last.

A model output is a starting point. The checks around it are what make the result useful in a document process.

  1. 01

    Extract as is

    Values are taken exactly as they appear in the document. Nothing is interpreted yet.

  2. 02

    Cross-check

    Where a document says the same thing twice, both are compared: an amount in digits and in words, a date here and a date there.

  3. 03

    Vote

    For documents with many layouts, several models read the same document independently. Agreement between them contributes to confidence.

  4. 04

    Confirm

    Every extracted value is searched for in the raw text of the document. A value that cannot be found there loses confidence.

  5. 05

    Normalise last

    Only at the end are dates, country codes, postcodes and names brought into a fixed format, giving downstream systems consistent values.

One confidence per field

  • 0.95 and up. Several methods agreed and the value was found in the document text.
  • Around 0.80. Most methods agreed. Suitable for most automated workflows.
  • 0.60 and below. Methods disagreed. Have a person look.
Who it is for

Advisers, intermediaries and lenders.

Advisers and intermediaries

Prepare a complete application faster, and see at intake which documents will not do: a black-and-white passport scan, a payslip from the wrong period, an unsigned declaration.

Lenders

Support checks, assessment and acceptance with fields you can compare against the application, each with a confidence you can set thresholds on.

The API processes one document at a time and returns what that document says. Comparing documents with each other and with the application stays in your process, where your business rules live.

Practicalities

A small, predictable API surface.

  • Hosted in the European Union.
  • One API key per client, sent as a header. Two keys are valid at any time, so a key rotation needs no maintenance window.
  • Results can be collected for one hour after processing; after that the API no longer returns them. That limits what a leaked key would be worth.
  • Volumes by agreement. A service level agreement is available on request.
Privacy and GDPR

Personal data, handled with care.

You stay in control

You remain the controller and we act as your processor, under a data processing agreement that is part of our terms. Ask us to delete your documents and they are deleted.

Only in the EEA

Documents are processed and stored in Microsoft Azure data centres within the European Economic Area. Microsoft is our only sub-processor.

Less exposed by design

API keys are stored only as hashes, each client sees only its own documents, and results can be collected for one hour. For an ID document the API reports whether the citizen service number is visible, never the number itself.

Clear agreements

Retention, deletion, security measures and data breach notification are set out in writing. We may use documents to improve the service, under strict conditions, and you can opt out with one email.

The terms and the data processing agreement apply to trials too. Read the terms and data processing agreement

Try it

Would you like to try it yourself?

Send us a message. We can provide a key and documentation, and help you run a small test with your own material.

info@smart-backoffice.nl

Our general terms and conditions, including the data processing agreement, apply to every use of Mortgage Documents, trials included. By submitting a document, you accept them on behalf of your organisation.

Read the terms